diff --git a/ares/ares.h b/ares/ares.h index 50ce7f850..03b6a924a 100644 --- a/ares/ares.h +++ b/ares/ares.h @@ -98,6 +98,8 @@ extern "C" { #define ARES_OPT_LOOKUPS (1 << 8) #define ARES_OPT_SOCK_STATE_CB (1 << 9) #define ARES_OPT_SORTLIST (1 << 10) +#define ARES_OPT_SOCK_SNDBUF (1 << 11) +#define ARES_OPT_SOCK_RCVBUF (1 << 12) /* Nameinfo flag values */ #define ARES_NI_NOFQDN (1 << 0) @@ -177,6 +179,8 @@ struct ares_options { int ndots; unsigned short udp_port; unsigned short tcp_port; + int socket_send_buffer_size; + int socket_receive_buffer_size; struct in_addr *servers; int nservers; char **domains; diff --git a/ares/ares_init.c b/ares/ares_init.c index 298dd7b90..6c36d6e2b 100644 --- a/ares/ares_init.c +++ b/ares/ares_init.c @@ -133,6 +133,8 @@ int ares_init_options(ares_channel *channelptr, struct ares_options *options, channel->ndots = -1; channel->udp_port = -1; channel->tcp_port = -1; + channel->socket_send_buffer_size = -1; + channel->socket_receive_buffer_size = -1; channel->nservers = -1; channel->ndomains = -1; channel->nsort = -1; @@ -320,6 +322,12 @@ static int init_by_options(ares_channel channel, channel->sock_state_cb = options->sock_state_cb; channel->sock_state_cb_data = options->sock_state_cb_data; } + if ((optmask & ARES_OPT_SOCK_SNDBUF) + && channel->socket_send_buffer_size == -1) + channel->socket_send_buffer_size = options->socket_send_buffer_size; + if ((optmask & ARES_OPT_SOCK_RCVBUF) + && channel->socket_receive_buffer_size == -1) + channel->socket_receive_buffer_size = options->socket_receive_buffer_size; /* Copy the servers, if given. */ if ((optmask & ARES_OPT_SERVERS) && channel->nservers == -1) diff --git a/ares/ares_process.c b/ares/ares_process.c index d359b115a..1db3d1129 100644 --- a/ares/ares_process.c +++ b/ares/ares_process.c @@ -25,6 +25,7 @@ #ifdef HAVE_SYS_UIO_H #include #endif +#include /* for TCP_NODELAY */ #include #include #include @@ -732,9 +733,36 @@ static int nonblock(ares_socket_t sockfd, /* operate on this */ #endif } +static int configure_socket(int s, ares_channel channel) +{ + nonblock(s, TRUE); + +#ifdef FD_CLOEXEC + /* Configure the socket fd as close-on-exec. */ + if (fcntl(s, F_SETFD, FD_CLOEXEC) == -1) + return -1; +#endif + + /* Set the socket's send and receive buffer sizes. */ + if ((channel->socket_send_buffer_size > 0) && + setsockopt(s, SOL_SOCKET, SO_SNDBUF, + &channel->socket_send_buffer_size, + sizeof(channel->socket_send_buffer_size)) == -1) + return -1; + + if ((channel->socket_receive_buffer_size > 0) && + setsockopt(s, SOL_SOCKET, SO_RCVBUF, + &channel->socket_receive_buffer_size, + sizeof(channel->socket_receive_buffer_size)) == -1) + return -1; + + return 0; + } + static int open_tcp_socket(ares_channel channel, struct server_state *server) { ares_socket_t s; + int opt; struct sockaddr_in sockin; /* Acquire a socket. */ @@ -742,8 +770,25 @@ static int open_tcp_socket(ares_channel channel, struct server_state *server) if (s == ARES_SOCKET_BAD) return -1; - /* Set the socket non-blocking. */ - nonblock(s, TRUE); + /* Configure it. */ + if (configure_socket(s, channel) < 0) + { + close(s); + return -1; + } + + /* + * Disable the Nagle algorithm (only relevant for TCP sockets, and thus not in + * configure_socket). In general, in DNS lookups we're pretty much interested + * in firing off a single request and then waiting for a reply, so batching + * isn't very interesting in general. + */ + opt = 1; + if (setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &opt, sizeof(opt)) == -1) + { + close(s); + return -1; + } /* Connect to the server. */ memset(&sockin, 0, sizeof(sockin)); @@ -777,7 +822,11 @@ static int open_udp_socket(ares_channel channel, struct server_state *server) return -1; /* Set the socket non-blocking. */ - nonblock(s, TRUE); + if (configure_socket(s, channel) < 0) + { + close(s); + return -1; + } /* Connect to the server. */ memset(&sockin, 0, sizeof(sockin));